FITNESS: Personalized Exercise Plans Using Health Data Insights

Paper: A Guideline-Grounded Multi-Agent Framework for Personalized Exercise Prescription from Clinical and Wearable Data

Authors: Xintong Li (Center for Data Science, Nell Hodgson Woodruff School of Nursing, Emory University, Department of Computer Science, Emory University), Yufen Lin (Nell Hodgson Woodruff School of Nursing, Emory University, Cancer Prevention and Control Research Program, Winship Cancer Institute, Emory University), Yuhua Wu (Center for Data Science, Nell Hodgson Woodruff School of Nursing, Emory University), Fengran Wang (Center for Data Science, Nell Hodgson Woodruff School of Nursing, Emory University), Minxiao Wang (Center for Data Science, Nell Hodgson Woodruff School of Nursing, Emory University), Rebecca A. Gary (Nell Hodgson Woodruff School of Nursing, Emory University), Jacob Kariuki (Lillian Carter Center for Global Health and Social Responsibility, Nell Hodgson Woodruff School of Nursing, Emory University), James Smith (Lancaster Medical School, Lancaster University), Sandra A. Billinger (Department of Neurology, University of Kansas Medical Center), Canhua Xiao (Nell Hodgson Woodruff School of Nursing, Emory University), Xiao Hu (Center for Data Science, Nell Hodgson Woodruff School of Nursing, Emory University, Department of Computer Science, Emory University, Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology and Emory University), Runze Yan (Center for Data Science, Nell Hodgson Woodruff School of Nursing, Emory University)

Executive Summary

The research introduces FITTER, a new framework designed to personalize exercise prescriptions for individuals based on their health data. This approach aims to streamline the process of translating general exercise recommendations into tailored plans that consider specific health conditions and personal circumstances, such as medical diagnoses, medications, and activity levels. The challenge has been scaling this personalization beyond specialized rehabilitation environments.

FITTER integrates various data sources, including clinical records and data from wearable devices, to formulate individualized exercise prescriptions. These prescriptions follow the FITT-VP model, which stands for Frequency, Intensity, Time, Type, Volume, and Progression. A notable feature of FITTER is its built-in safety audit and repair system, which ensures that the exercise recommendations comply with established safety constraints. This innovative safety mechanism addresses potential violations of guidelines, thereby enhancing the reliability of the prescriptions generated.

The evaluation of FITTER involved analyzing data from 400 participants in the All of Us research program, which included 300 individuals with cancer and 100 with stroke. The performance of FITTER was benchmarked against five other large language model (LLM) comparators using various computational methods and clinician reviews. The results demonstrated that FITTER achieved a remarkable 98.5% compliance with safety constraints, significantly outperforming the comparators, which ranged from 75.8% to 85.8% compliance. Additionally, FITTER exhibited superior precision in exercise dosing and alignment with established clinical guidelines.

The study also highlighted that the safety audit and repair component was the most significant factor contributing to FITTER's high compliance rates. Clinicians assessed 80 exercise prescriptions created by FITTER, focusing on aspects such as safety, personalization, clarity of instructions, expected effectiveness, and overall clinical acceptability. This clinician-facing approach ensures that the recommendations are practical and actionable, addressing the needs of both patients and healthcare providers.

The implications of this research are substantial. By effectively combining clinical and wearable data, FITTER could enhance the ability of healthcare professionals to prescribe exercise in a way that is safe and tailored to each patient's unique situation. This capability may lead to improved health outcomes for individuals with chronic conditions, particularly in settings where personalized care is critical. The framework may have the potential to be applied in various healthcare environments, though the current evaluation was conducted in a controlled research context rather than real-world clinical practice.

Academic Abstract

Translating population-level exercise guidance into individualized prescriptions requires integrating clinical and wearable data, including diagnoses, medications, functional limitations, and baseline activity, with condition-specific recommendations, a process difficult to scale beyond specialist rehabilitation settings. We developed FITTER, a multi-agent framework that integrates these data with condition-specific guidance to generate personalized FITT-VP prescriptions and applies deterministic safety audit and repair to correct rule-checkable violations. We evaluated FITTER using data from 400 All of Us participants (300 cancer, 100 stroke) against five LLM comparators through computational evaluation, ablation, and structured clinician review. FITTER achieved 98.5% safety-constraint compliance versus 75.8-85.8% for comparators and showed greater quantitative dosing precision and guideline alignment. Cancer-cohort ablation identified safety audit and repair as the largest contributor to safety-constraint compliance. Clinicians reviewed 80 prescriptions across safety, personalization, instruction concreteness, anticipated effectiveness, and clinical acceptability. FITTER provides a clinician-facing approach integrating clinical and wearable data, guidance, and explicit safety constraints.